Organizations deploying AI face a blind spot in their security infrastructure. A 2026 State of Agent Security Report found that roughly 1,280 third-party products now embed AI agents in typical enterprise environments. Of these, only 282 operate behind single sign-on systems, leaving approximately 1,000 agents invisible to identity management controls.

The visibility gap stems from how identity infrastructure operates. Identity stacks can only govern systems that authenticate through them. Most embedded AI agents bypass traditional authentication entirely, creating unmonitored access points within organizations.

This disconnect presents a security challenge. Organizations invest in protecting AI systems they deliberately deploy while remaining unaware of AI agents embedded in third-party tools already operating in their environments. The agents sitting outside identity infrastructure operate without the security controls, logging, or governance frameworks applied to intentionally deployed AI systems.

The problem extends beyond simple oversight. Third-party vendors embed AI into their products as standard functionality. When organizations adopt these products, they inherit the embedded agents whether security teams know about them or not. A vendor might update software to include AI capabilities without alerting customers to the change.

This creates a multilayered risk. Security teams cannot monitor what they cannot see. Agents outside identity systems lack centralized logging and audit trails. If these agents malfunction, get compromised, or access sensitive data, detection and response become significantly harder. The security infrastructure built for chosen AI systems provides no visibility into or control over these shadow agents.

Organizations cannot simply disable these agents through standard identity controls. Removing access requires either vendor cooperation or direct intervention at the product level. Many organizations lack the technical expertise or leverage to accomplish either.

The visibility gap persists because of architectural factors rather than intentional hiding. Identity systems authenticate humans and managed applications. Embedded AI agents typically authenticate differently, if they authenticate at all. They may operate with hardcoded credentials, API keys, or service accounts that bypass